typeset using JPSJ.sty <ver.l.Ob>
Optical Conductivity of the TwoDimensional Hubbard Model
Hiroki Nakano and Masatoshi Imada
Institute for Solid State Physics, University of Tokyo, Roppongi 7221, Minatoku, Tokyo 1068666
(Received )
Charge dynamics of the twodimensional Hubbard model is investigated. Lanczos
C^ \ diagonalization results for the optical conductivity and the Drude weight of this model are
presented. Near the Mott transition, large incoherence below the upperHubbard band is ob
KEYWORDS: twodimensional Hubbard model, Mott transition, optical conductivity, Drude weight, incoher
tained together with a remarkably suppressed Drude weight in two dimensions while the clearly
coherent character is shown in one dimension. The twodimensional results are consistent with
previous results from quantum Monte Carlo calculations indicating that the Mott transition
^ i in this twodimensional model belongs to the universality class characterized by the dynamical
^\^ , exponent of z = 4.
(N
O
0\ ' ence, exact diagonalization, dynamical exponent
^ ■ Physics of metal insulator transitions is one of the major issues in current condensedmatter
I ' physics from both experimental and theoretical viewpointsJH' The Hubbard model is one of the
Ch ' simplest models which can describe the metalinsulator transition. This model shows a Mott
o :
CJ . transition originating in strong correlation between electrons. Unfortunately, properties of this
model in two and three dimensions are not completely understood.
 ' — I I PI
/\ ' According to the scaling hypothesis for the Mott transition,a) the Drude weight and the com
C^ \ pressibility have the critical dependence on the doping concentration 5 as D cc (5^+(^~^)/'^ and
K oc 5^~^''^, respectively, where z denotes the dynamical exponent and d represents the spatial
dimension. The transition is characterized by z. The value of z may depend on systems and the
scaling theory (ST) itself does not determine it. To fix the value of z, one has to estimate it explic
itly, for example, using numerical results. Dagotto et al. reported D (x 6 for the twodimensional
(2D) Hubbard model.EP If one employs the ST, D cc 6 suggests z = 2, which is in the same
universality class as the band metalinsulator transition. From the quantum Monte Carlo (QMC)
calculations ,q' on the other hand, the chemical potential /i is scaled by 6'^. This leads to k oc 1/6
and z = 4. Another QMC calculation for the chemicalpotential dependence of the localization
length, ^i, shows ^i=\fJ' — fJ'c\~'^ with v = jp> which also suggests z = A. Thus, if one is based on
the ST, a discrepancy remains between these two QMC results and the above result by Dagotto et
al, which could pose a suspicion on the validity of the ST.
Quite recently, it was reported that D in the 2D tJ model has a very small value near the Mott
transition, which strongly indicates z = 4£P The critical behavior of k oc 1/5 in this model was
also reported.13) Since the tJ model is obtained in the strongcoupling limit of the Hubbard one
in small J/t, it is widely believed that these models show the same critical behavior at least at
realistic values of J/t. To have further insight on this issue, in our present analyses , we treat the
finitesize (FS) effect more carefully than in the literature and reexamine the optical conductivity
of the 2D Hubbard model.
We study a model given by 7^=— t X](J7),cr cJ^Cjo +f^Z]j^iT^ii' where the creation(annihilation)
of an electron at site i with spin a is denoted by c^(cjcr) with the number operator rijo Here, we
treat only the nearestneighbor hopping with amplitude t=l. The doping concentration is given
by 6=1— jf, where N^ and A'^g denote the number of electrons and sites, respectively. We choose a
boundary condition (BC) among periodic, antiperiodic and mixedEP ones so that the groundstate
energy becomes as low as possible as it should be. In fact, with this procedure, negative and
unphysical values of D obtained in Ref.^ near half filling do not appear because D is the curvature
of the flux dependent energy at the true ground state. The FS effect in this procedure is expected
to be smaller than the fixed BC. Actually, when one considers the sum of squared deviations from
the thermodynamic limit in D of 4x4site system with even N^ at ?7=0, the sum for the optimized
BC is ~0.0021, smaller than ~0.0032 for the BC fixed to be periodic.
We use exact diagonalization of an FS cluster of 4 x 4 sites based on Lanczos algorithm and
continuedfractionexpansion method.EP Hilbert space of this cluster has a huge dimension of order
of 10^. To treat the dimension, we carry out the parallel processing on supercomputers with a
distributedmemory system.
We here calculate the optical conductivity defined as o(w) = [(Tx{^) + <^j/('^)]/2 with (Ta{uj) given
by 2TTe^Da5{uj) + ^ En(^o) ^^l!f=^'^('^^".+Eo). Here D^ is the Drude weight along adirection
{a=x,y) and ja is a current operator along the adirection defined as ja = — iZ^i (T^(cL'^«+<5a,o —
ci 5 o'io")' where 5a is the unit vector along the adirection. \n) denotes an eigenstate with the
energy eigenvalue of En The ground state is represented by n = 0. The averaging operation is
performed to handle the anisotropic results due to the mixed BC and to reduce the FS correction.
The Drude weight can be obtained from the combination of (j{ijo) and the sum rule of /q°° a{uj)duj =
ne'^K, where 'IK denotes the kinetic energy per site. Hereafter we will call K the total weight. When
U is large enough, we also calculate an effective carrier density defined as N^s = ^ Jq" a{u>)duj,
where Wc is a frequency just below the upperHubbard (UH) band. If U is small, weight transfer
to the UH band and the one within the lowerHubbard band merge and A'efj is not well defined.
To find LOc definitely, we take a large value of U/t = 16. Note that, in the tJ limit of ?7— >oo,
Ncs is equal to K. Besides, one can expect that the large value of U would reduce the FS effect.
In this work, we make a further procedure to reduce the FS effect. The procedure is to multiply
a correction factor defined by r = Koo/Kn^, where K^o and K]\j^ are the total weights for U=0
in the thermodynamic Hmit and of the A'^gsite system, respectively. Hereafter, all the quantities
obtained after this procedure are labeled with suffix c, for example, D'^. To check the validity of our
calculations, we show results for the ID Hubbard chain with U/t = 16 in the inset of Fig. 1. It is
seen that our Drude weight excellently agrees with the Betheansatz result in the thermodynamic
limit.t3' Here one can also see the coherent feature, namely, A'^g — W^ <^ N^g in the whole region
of (5.
Now, we present the results for the 2D Hubbard model with U/t = 16 in Fig. 1. At half filling,
the Drude weight is positive and finite but very small. This small and finite value comes from
the FS effect and is expected to converge to zero if systems become larger. Actually, we have
obtained D'=~0.028 for ^lOx ^10 sites for the same U at half filhng whereas L>'=~0.008 for 4 x 4
sites, which shows a rapid convergence to zero. In the diluteelectrondensity region (6^1), the
interaction effect is small and the three quantities are close to the {7=0 case. At larger electron
density, the interaction works to prevent the electron motion and the three quantities decrease. In
the whole region of S, A'eff exhibits a smooth and convex curve. In contrast with the results in Ref. ^,
compared to N^s, D shows a concave shape below 6 = 0.25 as shown in the data point closest to
the Mott transition while an overall convex behavior for higher 6. This makes a quite asymmetric
curve of D between 6 < 0.5 and 5 > 0.5 under reflection with respect to 6=0.5. The concave shape
for small 5 suggests z > 2. Although the number of data point at small 5 is not sufficient to draw
a definite conclusion on the exponent, the present result obtained after carefully reducing the FS
effect supports that, as 5 — > 0, D vanishes faster than linear (5dependence, and does not contradict
D (X 6^. The exponent deduced from D cc 6'^ is consistent with the ST characterized by z = 4, and
agrees with other results from the QMC method.a'B)
Let us discuss the frequency dependence of the optical conductivity for ^^7^=16 shown in Fig. 2(a)
(d). At half filling, only the excitations to the UH band are seen in Fig. 2(a). Upon increasing 5,
weights are transferred to the region below the Hubbard gap. In Fig. 2(b), one can see the large
incoherent part below the UH band. The midgap incoherence has a larger weight on the lower
frequency side. As the frequency is increased, the height of the incoherence gradually decreases
after the highest peak at about u;/t~2. It is tempting to relate this structure to the shape of
the 1/uj tail seen in the experiment of highTc compound, although the incoherent response in the
small frequency side uj/t <2 clearly suffers from the FS effect. A similar behavior is seen in the
study of the 2D tJ model.til'c3'Ej) Note that the data which Dagotto et al. reported in Ref. ^ did
not show such structure because the midgap incoherence itself is too small at the same 5=0.125.
Note also that the present calculations show that the incoherent peak gradually loses its weight
with increasing 5 while the longtail structure survives.
In summary, we have investigated the charge dynamics of the 2D Hubbard model. Lanczos
diagonalization results under the careful treatment of the boundary condition have been presented.
Near the metalinsulator transition, the optical conductivity has the large incoherence in the mid
gap region. At that doping concentration, the Drude weight is clearly suppressed, which supports
the z = 4 universality class of the Mott transition.
The authors thank H. Tsunetsugu and Y. Motome for fruitful discussions. This work is supported
by 'Research for the Future Program' from Japan Society for Promotion of Science (JSPSRFTF
97P01103). Computations was partly performed using the facilities of the Supercomputer Center,
Institute for Solid State Physics (ISSP), University of Tokyo. Parallel calculations were done at
FUJITSU VPP500 in ISSP and VPP700 in Kyushu University.
[1] M. Imada, A. Fujimori and Y. Tokura: Rev. Mod. Phys. 70 (1998) 1039, and references therein.
[2] M. Imada: J. Phys. Soc. Jpn. 64 (1995) 2954.
[3] E. Dagotto, A. Moreo, F. Ortolani, D. Poiblanc and J. Riera: Phys. Rev. B 45 (1992) 10741. In this resuh, due
to the finitesize effect, the Drude weight at half fiUing calculated under the periodic BC is unphysically and
largely negative. In addition, unfortunately, they do not show data ai 5 — 0.125, which are not on the linear line
A^cff oc 5.
[4] N. Furukawa and M. Imada: J. Phys. Soc. Jpn. 61 (1992) 3331; ibtd. 62 (1993) 2557.
[5] F. Assaad and M. Imada: Phys. Rev. Lett. 76 (1996) 3176.
[6] H. Tsunetsugu and M. Imada: J. Phys. Soc. Jpn. 67 (1998) 1864.
[7] M. Kohno: Phys. Rev. B 55 (1997) 1435.
[8] A mixed BC in 2D is composed of the antiperiodicity along a;direction and the periodicity along j/direction,
or the opposite combination. It corresponds to the case when the AharonovBohm fiux $ in the hopping term
introduced as the twisted BC is (7r,0) or (0, vr). In Ref. hll it was reported that there exist cases where <l>=(7r,0)
or (0, tt) gives the lowest energy in the 2D tJ model.
[9] E. Gagliano and C.Balseiro: Phys. Rev. Lett. 59 (1987) 2999.
[10] N. Kawakami and S. K. Yang: Phys. Rev. B 44 (1991) 7844.
[11] D. Poilblanc: Phys. Rev. B 44 (1991) 9562.
[12] E. D. Dagotto: Rev. Mod. Phys. 66 (1994) 763.
[13] J. Jaklic and P. Prelovsek: Phys. Rev. B 52 (1995) 6903.
Fig. 1. Drude weight, effective carrier density and total weight of the 2D Hubbard model when U/t = 16. Inset
shows the ID results for 10 and 12 sites for U/t = 16 together with the Betheansatz result in the thermodynamic
hmitw' (solid curve) for comparison.
Fig. 2. Incoherent part of cr(a;) for (a) 5=0 (antiperiodic BC), (b) 5=0.125 (mixed), (c) 5=0.25 (mixed), (d) 5=0.375
(periodic). Delta functions in the peaks are broadened with width of e=0.05.
U/t=0, N = oo
0.6
— 1 ' '
U/t=0, N = oc
0.0
0.2
0.4 0.6
5
0.8
1.0
0.12
.^0.10
0.08
§0.06
o
'gO.04
0.02
0.00
(a)60
J
\jM
w
1 . .
. IwuIUaaa.. . ■
0.06
Oo.oi
a,
oo.oi
(b)50.125
10 15
(O/t
20
25
0.20
'■■■■'■■■■
>^
(d) 50.375
■^0.15
o
O 0.10
o


o
g^O.05
o
. . 1 JuA*,aa.
n na
. J
WLJH....A
10 15
(O/t
20
25